Embedding-based Field Matching

Leverage embeddings of both user utterances and annotated JSON metadata to search for matching criteria and use LLM confidence thresholds to decide which fields to populate.